It's an interesting discussion. It almost seems that there is a 'gap in the market' so to speak. The gap being the application of accurate algorithms to enterprise search applications to improve the accuracy of the search results. As you say a query of "Well-1" won't return the same results as "Well 1". I guess the challenge is the shear volume of information we are dealing with and the willingness of companies to invest the time and money to improve their enterprise search capabilities.
Original Message:
Sent: 08-28-2015 05:46
From: Paul Cleverley
Subject: Causes for poor enterprise search tasks
Hi Dennie
Good question and indeed a common question. Internet search phenomena (such as Google) are undoubtedly much better than their enterprise search cousins in terms of helping people find what they need for many reasons. Putting search literacy aside, some thoughts here include:
More forgiving algorithms
In many enterprise search engines, syntax wise, a query of ‘Well-1’ will often not return results that I will find making a query with ‘Well 1’. Likewise ‘health centre’ may miss the key web page titled ‘health center’. From my experience, the algorithms to sort out misspellings (using techniques like Levenshtein distance on an n-gram index) can be poor in enterprise search out of the box deployments. Many enterprise search deployments do not employ sophisticated ‘query dropping or re-formulation’ (not just relating to common stop words) like Internet search engines. For example, a query such as ‘health and safety incident register template guide’, may miss the key item ‘health and safety incident register template’ simply because an unforgiving Boolean AND is being made between all the query terms by the enterprise search engine and the key item simply does not contain the word ‘guide’, the searcher has been penalised for adding 'too many words' (not desirable).
I also feel that when enterprises combine web and document content, search ranking algorithms are not always tuned properly. If someone make a query with one word (60-70% by volume of most enterprise search queries) in my view they are unlikely to be looking for a document (more so information on an Intranet/Wiki web page), so we can probably be smarter with our ranking algorithms.
Bear in mind things are not all rosy on the Internet. Many geoscientists are shocked when I ask them to use Google Scholar to find reports on Peru Magnetics. If they did not also search on the singular ‘Peru Magnetic’ (without the ‘s’ on the end) they would have missed 97% of the relevant content!!!!
From a semantics perspective, I recall a study some time ago of 4Million articles in the Le Monde newspaper. The terms ‘Web’ and ‘Internet’ never occurred near one another. However, using word co-occurrence algorithms (the words used before and after these terms), it was possible for automated algorithms to infer these words virtually mean ‘the same concept’ from their surrounding associated words (context), helping ‘normalize’ search queries. Internet search engines deploy these techniques (such as latent semantic indexing) to help mitigate language issues and also use English thesauri (such as Wordnet) to understand word variants. Enterprises should take advantage of these techniques, but we need to consider we have more technical language, less statistical information and plenty of corporate acronyms. No matter what the search engine, unless we teach it we will probably struggle to locate certain information and cope with the ambiguities of language.
Statistical usage data challenge
You may have heard of ‘big data trumps smarter algorithms’ and this can be applied to statistical usage data. Internet usage of some search engines hits 1Billion queries a week. This statistical (crowdsourced) information can be used to help promote the most relevant sites (using this ‘click-through’ data) as dictated by ‘the people’. Outside a handful of the most commonly made queries in an enterprise (e.g. IT Request), the statistical data in enterprise search is unlikely to be anywhere near as useful, regardless of what search engine technology you choose. Many enterprise searchers may be making queries that nobody else has ever made in the organization (or so few times, statistical information is not helpful).
Having said that, some commentators are concerned that for some types of information on the Internet people are increasingly finding information through the ‘rear view mirror’, where ranking is determined by what people ‘have looked at’ and can become self-perpetuating, especially for ‘news’.
Content availability
If I make an Internet query such as ‘Finding odd one out in two excel columns’, I find what I need (discussion forums normally serve these common ‘how do I’ questions). The search engine has helped, but more importantly the reason my need was met is because the content is there on the behemoth that is the Internet. I often see people struggling to answer a question in the enterprise search (or find something), when the information or answer is simply not written down or contained within the index of the search engine. People blame the search, but it’s really a content issue. Enterprises also have to handle file permissions, which is not an issue on the Internet.
User Context (information architecture)
To cope with information overload, context is used like a ‘magnet’ to sift out the needles from the information haystack. Internet search engines have country servers, so have an immediate advantage of localizing some content (based on what location you are in) and use cookies to store browsing histories. The majority of enterprise search deployments could use context more. It is not easy to broadly apply context filters effectively (risk of hiding information from people that they may find useful), but invariably more could be done in the enterprise.
Perhaps we should think of enterprise search as a range of user interfaces in certain domains (each one tailored to the necessary context and functionality to meet a suite of work tasks) not just a single monolithic ‘search page’.
Enterprise search has many challenges, more forgiving algorithms can help, but are probably just one part of an approach to achieving higher levels of satisfaction and search task outcome performance.
Cheers, Paul
------------------------------
Paul Cleverley
Robert Gordon University
------------------------------
Original Message:
Sent: 08-27-2015 13:44
From: Dennie Heye
Subject: Causes for poor enterprise search tasks
Hi Paul,
I have done many search awareness sessions in the past, and many people indeed had poor information searching skills. Especially the conversion from a problem to a search query - which requires effort and creativity. People would complain that this was too hard and that "public" search engines had better technology that resolved their poor searching technique. Was is your take on this?
Best regards,
Dennie
------------------------------
Dennie Heye
Business Analyst Information Management
Royal Dutch Shell
------------------------------
Original Message:
Sent: 08-23-2015 15:41
From: Paul Cleverley
Subject: Causes for poor enterprise search tasks
Dear colleagues
Some recent research that may be of interest. A deep dive investigation of data collected from the feedback button on an enterprise search user interface in a large multinational organization yields some interesting results. Analysis of 891 comments collected over two years, indicates that Technical/IT issues (such as reliability and ranking) and Information Management issues (e.g. publishing of content and naming) are causal factors for poor search experiences. What has received less attention in the literature is the role that search literacy plays in poor search experiences. In this study, 22% of all poor search experiences were caused by literacy factors. In particular, users unable to form appropriate queries based on the information need in hand. It is often said that 'you should not need to train users to search if the search tool is good enough'. But is that really the case? More at: www.paulhcleverley.com
Regards, Paul
------------------------------
Paul Cleverley
Robert Gordon University
Email: p.h.cleverley@...
------------------------------